Output Loop Bench · Reasoning

Prospection challenge: AI models compared

Insight from the prospection database, the same questions each run. 9 models tested on the same task.

Yes/no facts per business

544 local trade businesses (landscapers, electricians, removal companies, roofers, drainage companies and plumbers) from the original research. Each decision model answers seven yes/no questions about every business, and the Lab adds the answers up. The top row is the correct answer, computed from the same rules, so every miss shows as a difference.

ModelMakerTimeTokens
JevTypeSafe38.7s1,740,547
Perplexity DecisionsPerplexity77.4s1,603,663
gemini-4-argon-mediumGoogle421.5s508,490
kimi-k3-maxMoonshot (Kimi)480.5s543,606
claude-haiku-5-5Anthropic660.5s2,748,961
ClefOther793.6s572,825
gpt-6-lunaOpenAI875.2s520,444
glm-5.3Z.ai1108.6s565,868
qwen3.8-flashAlibaba (Qwen)1395.4s611,740

Open Prospection challenge in the interactive Bench